Two AI Deals Reveal a New Phase of U.S.-China Tech Relations
Two low-profile deals struck in the shadow of DeepSeek V4 Pro's launch are quietly redrawing the fault lines of the global AI industry — revealing that the most consequential shift in U.S.-China tech relations is no longer about competition or regulatory détente, but about cold commercial interdependence.
The first deal: ByteDance has emerged as one of Microsoft's largest AI clients globally, with its annual spending on Microsoft Azure and AI services tracking toward $1 billion, according to people familiar with the matter — a revenue stream significant enough that it factored directly into Microsoft's decision, made several years ago, to retain its China operations rather than exit the market. The second: International Business Machines (IBM) has signed a multi-year, $240 million agreement with Together AI, a U.S. AI infrastructure company whose Model-as-a-Service platform supports Chinese open-source models including DeepSeek, MiniMax, Kimi, and GLM — a move that could funnel Chinese AI into the procurement pipelines of the world's most conservative institutional buyers.
Taken together, the two developments signal a structural phase change: after years of direct competition and awkward joint-venture arrangements, U.S. and Chinese technology companies are converging on a new division of labor — one organized around AI, global market expansion, and mutually reinforcing commercial interest.
China Revenue Rescues Microsoft's Fading China Franchise
Microsoft's China business had, by most internal and external measures, become a liability. China accounted for only approximately 1.5% of Microsoft's total revenue in fiscal year 2024, a figure that underscores how thoroughly the company had been outmaneuvered in its core consumer and enterprise segments. Azure held roughly 10% of China's public cloud market — a distant second tier behind Alibaba Cloud, Tencent Cloud, and Huawei Cloud. Windows 11 remained the flagship consumer product, bundled with a Copilot assistant widely regarded as underpowered by Chinese market standards. In an environment where Doubao, Qwen, and WorkBuddy were redefining productivity software with native AI capabilities, Microsoft's China-facing product lineup had grown conspicuously dated.
The company had reportedly considered a full exit from the Chinese market in the early 2020s.
What reversed that calculus was not a product revival. It was an outbound revenue model. Judson Althoff, then Microsoft's chief commercial officer, told an internal sales meeting in July 2025 that China had become the fastest-growing region for Microsoft's AI revenue. Azure AI revenue in China grew 400% in calendar year 2024, and roughly doubled again in the fiscal year ending June 2025 — outpacing every other sales region globally.
The client roster reads like a directory of China's most globally ambitious corporations: ByteDance, Ant Group, Meituan, and SHEIN, among others. The common thread is overseas expansion. These companies are not buying Azure to serve Chinese consumers — they are buying it to power international operations, leveraging Microsoft's network of data centers across more than 60 regions, its comprehensive global compliance certification stack, and its privileged access to OpenAI's latest model releases, including the GPT-5.6 series launched in early July 2026.
For Chinese companies building international content moderation, multilingual advertising targeting, and cross-border customer service platforms, GPT-series models retain a meaningful edge in non-Chinese language comprehension and cultural context — making the Azure OpenAI Service a pragmatic infrastructure choice rather than a prestige purchase.
Insiders describe the business model as structurally superior to Microsoft's legacy China operations: high-margin, asset-light, sticky, and scalable. By the mid-2020s, helping Chinese enterprises expand globally had become Microsoft China's largest single business line. ByteDance alone is approaching $1 billion in annualized Azure spend — a figure that, while still small relative to Microsoft's global revenue base of over $270 billion, represents a credible growth vector with compounding dynamics as Chinese companies deepen their international footprints.
IBM Lends Its Enterprise Credibility to Bridge DeepSeek's Trust Gap
The structural barrier facing Chinese open-source AI models in global enterprise markets is not performance. It is institutional trust.
Data compiled jointly by OpenRouter and venture firm Andreessen Horowitz (a16z) illustrates the trajectory: Chinese open-source models accounted for just 1.2% of weekly token consumption on tracked platforms at the end of 2024. By late 2025, that share had touched nearly 30% in peak weeks. Entering 2026, Chinese models have stabilized above 30% of weekly enterprise client traffic in the United States, with occasional spikes to 46%. DeepSeek alone commands approximately 17% share, making it the single largest open-source model provider by usage volume.
Yet that traction is concentrated among developers, startups, and mid-market technology firms. The procurement processes of Fortune 500 financial institutions, hospital networks, and government contractors operate in an entirely different register. Chief information officers at major banks and insurers do not download model weights from Hugging Face. Their approved vendor lists contain names like IBM, Microsoft, Amazon Web Services, and Google — and adding a new name to that list requires years of relationship-building, security auditing, and regulatory validation that no Chinese AI company has yet completed independently.
IBM is moving to close that gap, and the mechanism is characteristically indirect.
In February 2025, IBM added two distilled models based on DeepSeek-R1, along with support for custom imports of DeepSeek-R1's Qwen distillation variants, to the on-demand deployment catalog of its enterprise AI platform watsonx.ai — marking the first time a Chinese open-source reasoning model had entered the official product catalog of a top-tier global enterprise IT vendor.
In August 2026, IBM extended that commitment materially, signing a $240 million multi-year agreement with Together AI. The deal includes plans to deploy Nvidia HGX B300 compute clusters on IBM Cloud infrastructure, with an expected go-live date in the first quarter of 2027. Together AI's platform supports DeepSeek, MiniMax, Kimi, and GLM, among other models — meaning IBM is effectively building a distribution channel for Chinese open-source AI into its enterprise client base.
IBM has been careful to note that DeepSeek models listed on watsonx.ai are "not IBM models and carry no IBM warranty." But that disclaimer does not neutralize the implicit signal. When a model appears in IBM's product matrix, the message received by enterprise buyers is that IBM has conducted due diligence, is prepared to provide governance tooling and technical support, and stands behind the integration at an operational level. For regulated industries where vendor accountability is non-negotiable, that implicit endorsement carries weight that no benchmark score can replicate.
IBM's client base spans virtually every heavily regulated sector globally — commercial banking, insurance, healthcare, defense contracting, and government — making it the only technology company positioned to route Chinese AI into all of them simultaneously.
A Third Phase Emerges in U.S.-China Tech Relations
The ByteDance-Microsoft and DeepSeek-IBM dynamics are not isolated transactions. They represent the maturation of a structural pattern that has been building since the early 2020s.
The first phase of U.S.-China tech interaction — running roughly from the late 1990s through the mid-2010s — was defined by direct competition on Chinese soil. Google versus Baidu in search; Amazon versus Alibaba and JD.com in e-commerce; Uber versus Didi in ride-hailing. American companies entered with capital and brand advantages; Chinese companies won on local knowledge, regulatory navigation, and product iteration speed. The outcome was near-total displacement of foreign players from consumer-facing markets.
The second phase, beginning around 2015, was defined by structured partnership and localization. Microsoft partnered with 21Vianet to operate Azure in China; Amazon Web Services aligned with Beijing Sinnet Technology; Uber sold its China business to Didi in exchange for equity. These arrangements acknowledged the limits of direct competition but generated their own frictions — misaligned incentives, governance complexity, and diminishing strategic returns.
The current phase is categorically different. Chinese and American companies are no longer fighting over the same domestic market or managing the tensions of a shared legal entity. They are dividing the global market along lines of comparative advantage: Chinese companies bring scale, application-layer innovation, cost efficiency, and manufacturing ecosystem depth; American companies bring global infrastructure, regulatory credentialing, enterprise distribution, and foundational model capability. The target market is the world, not China.
Ma Wei, an assistant research fellow at the Institute for American Studies of the Chinese Academy of Social Sciences, characterizes the dynamic as structurally complementary: the United States holds advantages in foundational research, elite talent, core algorithms, and compute ecosystems, while China leads in application scenarios, industrial integration, engineering iteration velocity, and large-scale market deployment. Those profiles do not overlap — they interlock.
Apple, Google, and Tesla have each deepened China partnerships in parallel, reinforcing the pattern across sectors and companies.
The commercial logic is self-reinforcing. ByteDance's overseas expansion generates Azure revenue that justifies Microsoft's China presence; Microsoft's China presence generates the local talent and client relationships that feed the next round of deals. IBM's DeepSeek integration generates enterprise adoption data that strengthens DeepSeek's credibility with the next tier of institutional buyers; that credibility generates demand for more IBM-hosted capacity. Each transaction makes the next one more likely.
The AI wave has not merely created new products. It has created a new organizational grammar for how the world's two largest technology ecosystems relate to each other — one defined less by ideology or geopolitics than by the arithmetic of global market share.
Related Coverage:
ByteDance's AI Pivot: Why China's Tech Giant Is Betting Its Future on Enterprise Productivity
DeepSeek Open-Sources Harness Agent Runtime, Targeting the AI Execution Layer